Output shape of Conv1D

[Changbin_Du]

In C4_W4_Lab_1_LSTM.ipynb,
When build the model,

Reset states generated by Keras

tf.keras.backend.clear_session()

Build the model

model = tf.keras.models.Sequential([
tf.keras.layers.Conv1D(filters=64, kernel_size=3,
strides=1, padding=“causal”,
activation=“relu”,
input_shape=[window_size, 1]),
tf.keras.layers.LSTM(64, return_sequences=True),
tf.keras.layers.LSTM(64),
tf.keras.layers.Dense(1),
tf.keras.layers.Lambda(lambda x: x * 400)
])

Print the model summary

model.summary()

Why the output of conv1d (Conv1D) is (None, 20, 64) ?
How to understand the new_step calculation? Thanks so much!

[balaji.ambresh]

Here’s the explanation:

1. None: This refers to the batch size. Since this unknown till the time you invoke fit on the model, None serves as a placeholder.
2. 20: Here’s the explanation for causal padding.
3. 64: This is equal to the number of filters you use in the Conv1D layer

[Changbin_Du]

Thanks so much, Balaji!

My understanding is, with causal padding, the output shape will be same as input shape. So the new_step = original step, is this correct?

[balaji.ambresh]

I don’t understand what step means. When stride = 1, the same number of outputs are generated as the input window width.

[Changbin_Du]

Conv1D
The input is (batch_size, steps, input_dim), the steps=window_size.
output shape is (batch_size, new_steps, filters)
new_steps=steps=window_size if padding is causal.

[balaji.ambresh]

We both are saying the same thing. Thanks for clarifying.